|Publication number||US7039824 B2|
|Application number||US 10/918,209|
|Publication date||May 2, 2006|
|Filing date||Aug 13, 2004|
|Priority date||Feb 28, 2001|
|Also published as||US6779123, US20020120880, US20050039066|
|Publication number||10918209, 918209, US 7039824 B2, US 7039824B2, US-B2-7039824, US7039824 B2, US7039824B2|
|Inventors||Thomas D. Simon, Rajeevan Amirtharajah|
|Original Assignee||Intel Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (5), Non-Patent Citations (1), Referenced by (6), Classifications (10), Legal Events (4)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is a divisional (and claims the benefit of priority under 35 USC 120) of U.S. application Ser. No. 09/796,344, filed Feb. 28, 2001 now U.S. Pat. No. 6,779,123.
This invention relates to calibrating return time.
Digital electronics systems, such as computers, must move data among their component devices at increasing rates to take full advantage of the higher speeds at which these component devices operate. For example, a computer may include one or more processors that operate at frequencies of a gigahertz (GHz) or more. The data throughput of these processors outstrips the data delivery bandwidth of conventional systems by significant margins.
The digital bandwidth (BW) of a communication channel may be represented as:
Here, FS is the frequency at which symbols are transmitted on a channel and NS is the number of bits transmitted per symbol per clock cycle (“symbol density”). Channel refers to a basic unit of communication, for example a board trace in single ended signaling or the two complementary traces in differential signaling.
Conventional strategies for improving BW have focused on increasing one or both of the parameters FS and NS. However, these parameters cannot be increased without limit. For example, a bus trace behaves like a transmission line for frequencies at which the signal wavelength becomes comparable to the bus dimensions. In this high frequency regime, the electrical properties of the bus must be carefully managed. This is particularly true in standard multi-drop bus systems, which include three or more devices that are electrically connected to each bus trace through parallel stubs.
Practical BW limits are also created by interactions between the BW parameters, particularly at high frequencies. For example, the greater self-induced noise associated with high frequency signaling limits the reliability with which signals can be resolved. This limits the opportunity for employing higher symbol densities.
Modulation techniques have been employed in some digital systems to encode multiple bits in each transmitted symbol, thereby increasing NS. Use of these techniques has been largely limited to point-to-point communication systems, particularly at high signaling frequencies. Because of their higher data densities, encoded symbols can be reliably resolved only in relatively low noise environments. Transmission line effects limit the use of modulation in high frequency communications, especially in multi-drop environments.
Information communicated over buses, like information communicated over other transmission lines, may experience delay with reference to a system clock or other timing mechanism.
The present invention may be understood with reference to the following drawings, in which like elements are indicated by like numbers. These drawings are provided to illustrate selected embodiments of the present invention and are not intended to limit the scope of the invention.
A mechanism for calibrating return time is described here in the context of a high bandwidth communication technique that provides greater control over the frequency and encoding mechanisms employed to transfer data.
In one aspect of the invention, clock calibration information is stored with respect to each of at least two slave devices with which a master device will communicate using a bus, the clock calibration information being based on clock signals local to the master device and return clock signals corresponding to each of the slave devices. After the clock calibration information has been stored, data signals that are received from each of the slave devices are resynchronized based on the corresponding stored clock calibration information.
Before proceeding to a discussion of the calibration technique, we describe an example of the circuitry to which it may be applied.
For one embodiment of the invention, multi-drop bus system 200 is a computer system and devices 220 correspond to various system components, such as processors, memory modules, system logic and the like.
In the following discussion, various time-domain modulation schemes are used for purposes of illustration. The benefits of the present invention are not limited to the disclosed modulation schemes. Other time-domain modulation schemes, such as shape modulation (varying the number of edges in a pulse), narrowband and wideband frequency-domain modulation schemes, such as frequency modulation, phase modulation, and spread spectrum, or combinations of both time and frequency-domain modulation schemes (a pulse superposed with a high frequency sinusoid), are also suitable for use with this invention.
In the following discussion, a “pulse” refers to a signal waveform having both a rising edge and a falling edge. For pulse-based signaling, information may be encoded, for example, in the edge positions, edge shapes (slopes), and signal amplitudes between edge pairs. The present invention is not limited to pulse-based signaling, however. Other signal waveforms, such as edge-based signaling and various types of amplitude, phase, or frequency-modulated periodic waveforms may be implemented as well. The following discussion focuses on modulation of pulse-based signaling schemes to elucidate various aspects of the present invention, but these schemes are not necessary to practice the invention. Considerations similar to those discussed below for pulse-based signaling may be applied to other signal waveforms to select an appropriate modulation scheme.
For signal 410, the value of a first bit (0 or 1) is indicated by where (p0 or p1) the leading edge of symbol 420 occurs in the symbol period (phase modulation or PM). The values of second and third bits are indicated by which of four possible widths (w0, w1, w2, w3) the pulse has (pulse-width modulation or PWM). The value of a fourth bit is indicated by whether the falling edge has a large (rt0) or small (rt1) slope (rise-time modulation or RTM), and the value of a fifth bit is indicated by whether the pulse amplitude is positive or negative (a0, a1) (amplitude modulation or AM). Bold lines indicate an actual state of symbol 420, and dashed lines indicate other available states for the described encoding schemes. A strobe is indicated within the symbol period to provide a reference time with which the positions of the rising and falling edges may be compared. The number of bits encoded by each of the above-described modulation schemes is provided solely for illustration. In addition, RTM may be applied to the rising and/or falling edges of symbol 420, and AM may encode bits in the magnitude and/or sign of symbol 420.
PM, PWM, and RTM are examples of time-domain modulation schemes. Each time-domain modulation scheme encodes one or more bits in the time(s) at which one or more events, such as a rising edge or a rising edge followed by a falling edge, occur in the symbol period. That is, different bit states are represented by different event times or differences between event times in the symbol period. A bit interval associated with each time-domain modulation scheme represents a minimum amount of time necessary to reliably distinguish between the different bit states of the scheme. The modulation schemes selected for a particular system, and the number of bits represented by a selected modulation scheme is determined, in part, by the bit intervals of the candidate modulation schemes and the time available to accommodate them, i.e. the symbol period.
Similarly, one bit interval of duration t3 is allocated within the symbol period to allow the two states (rt0, rt1) to be distinguished reliably. The size of t3 is determined by noise and circuit limitations associated with rise time measurements. For example, rise times are differentiated by passing through coupler 240. Consequently, t3 must be long enough to allow the measurement of a second derivative.
Three bit intervals of duration t2 are allocated within the symbol period to allow the four states (w0, w1, w2, w3) to be reliably distinguished. The size of t2 is determined by noise and circuit limitations associated with pulse width measurements. If pulse width is determined relative to a clock strobe, considerations regarding clock jitter may apply. If pulse width is determined relative to, e.g., the leading edge of a pulse, considerations such as supply voltage variations between the measurements of the leading and trailing edges may apply.
In general, the time needed to encode an n-bit value in a time-domain modulation scheme (i) that has a bit interval, ti, is (2n−1)Ěti. If non-uniform bit intervals are preferred for noise or circuit reasons, the total time allotted to a modulation scheme is the sum of all of its bit intervals. When multiple time-domain modulation schemes are employed, the symbol period should be long enough to accommodate Σ(2n(i)−1)Ěti, plus any additional timing margins. Here, the summation is over all time-domain modulation schemes used. In the above example, the symbol period should accommodate t1+t3+3t2, plus any other margins or timings. These may include minimum pulse widths indicated by channel bandwidth, residual noise, and the like.
Using multiple encoding schemes reduces the constraints on the symbol time. For example, encoding five bits using pulse width modulation alone requires at least 31Ět2. If t2 is large enough, the use of the single encoding scheme might require a larger symbol period (lower symbol frequency) than would otherwise be necessary.
A minimum resolution time can also be associated with amplitude modulation. Unlike the time domain modulation schemes, amplitude modulation encodes data in pulse properties that are substantially orthogonal to edge positions. Consequently, it need not add directly to the total bit intervals accommodated by the symbol period. For example, amplitude modulation uses the sign or magnitude of a voltage level to encode data.
The different modulation schemes are not completely orthogonal, however. In the above example, two amplitude states (positive and negative) encode one bit, and the minimum time associated with this interval may be determined, for example, by the response time of a detector circuit to a voltage having amplitude, A. The pulse width should be at least long enough for the sign of A to be determined. Similarly, a symbol characterized by rise-time state rt1 and width state w3 may interfere with a next symbol characterized by phase state p0. Thus, noise and circuit limitations (partly summarized in the bit intervals), the relative independence of modulation schemes, and various other factors are considered when selecting modulation schemes to be used with the present invention.
The disclosed embodiment of interface 230 includes a transceiver 510 and a calibration circuit 520. Also shown in
Transceiver 510 includes a receiver 530 and a transmitter 540. Receiver 530 recovers the bits encoded in the transferred waveform on device side component 242 of electromagnetic coupler 240 and provides the recovered bits to the device associated with interface 230. Embodiments of receiver 530 may include an amplifier to offset the attenuation of signal energy on transmission across electromagnetic coupler 240. Transmitter 540 encodes data bits provided by the associated device into a symbol and drives the symbol onto device side 242 of electromagnetic coupler 240.
Calibration circuit 520 manages various parameters that may impact the performance of transceiver 510. For one embodiment of interface 230, calibration circuit 520 may be used to adjust termination resistances, amplifier gains, or signal delays in transceiver 510, responsive to variations in process, temperature, voltage, and the like.
For the disclosed embodiment of transceiver 510, transmitter 540 includes a phase modulator 640, a pulse-width modulator 630, an amplitude modulator 620 and an output buffer 610. Output buffer 610 provides inverted and non-inverted outputs to pads 602 and 604, respectively, to support differential signaling. A clock signal is provided to phase modulator 640 to synchronize transceiver 510 with a system clock. The disclosed configuration of modulators 620, 630, and 640 is provided only for purposes of illustration. The corresponding modulation schemes may be applied in a different order or two or more schemes may be applied in parallel.
The disclosed embodiment of receiver 530 includes an amplifier 650, an amplitude demodulator 660, a phase demodulator 670, and a pulse-width demodulator 680. The order of demodulators 660, 670, and 680 is provided for illustration and is not required to implement the present invention. For example, various demodulators may operate on a signal in parallel or in an order different from that indicated.
Devices 690(a) and 690(b) (generically, “device 690”) act as on-chip termination impedances, which in one embodiment of this invention are active while interface 230 is receiving. The effectiveness of device 690 in the face of, e.g., process, temperature, and voltage variations may be aided by calibration circuit 520. For transceiver 600, device 690 is shown as an N device, but the desired functionality may be provided by multiple N and/or P devices in series or in parallel. The control provided by calibration circuit 520 may be in digital or analog form, and may be conditioned with an output enable.
The disclosed embodiment of transmitter 540 modulates a clock signal (CLK_PULSE) to encode four outbound bits per symbol period. One bit is encoded in the symbol's phase (phase bit), two bits are encoded in the symbol's width (width bits) and one bit is encoded in the symbol's amplitude (amplitude bit). Transmitter 540 may be used to generate a differential symbol pulse per symbol period, and strobe transmitter 790 may be used to generate a differential clock pulse per symbol period.
Phase modulator 640 includes a MUX 710 and delay module (DM) 712. MUX 710 receives a delayed version of CLK_PULSE via DM 712 and an undelayed version of CLK_PULSE from input 704. The control input of MUX 710 transmits a delayed or undelayed first edge of CLK_PULSE responsive to the value of the phase bit. In general, a phase modulator 640 that encodes p phase bits may select one of 2p versions of CLK_PULSE subject to different delays. For the disclosed embodiment, the output of phase modulator 640 indicates the leading edge of symbol 420 and serves as a timing reference for generation of the trailing edge by width modulator 630. A delay-matching block (DMB) 714 is provided to offset circuit delays in width modulator 630 (such as the delay of MUX 720) which might detrimentally impact the width of symbol 420. The output of DMB 714 is a start signal (START), which is provided to amplitude modulator 620 for additional processing.
Width modulator 630 includes DMs 722, 724, 726, 728, and MUX 720 to generate a second edge that is delayed relative to the first edge by an amount indicated by the width bits. The delayed second edge forms a stop signal (_STOP) that is input to amplitude modulator 620 for additional processing. For the disclosed embodiment of transmitter 540, two bits applied to the control input of MUX 720 select one of four different delays for the second edge, which is provided at the output of MUX 720. Inputs a, b, c, and d of MUX 720 sample the input signal, i.e. the first edge, following its passage through DMs 722, 724, 726, and 728, respectively. If the width bits indicate input c, for example, the second edge output by MUX 720 is delayed by DM 722+DM 724+DM 726 relative to the first edge.
Amplitude modulator 620 uses START and _STOP to generate a symbol pulse having a first edge, a width, and a polarity indicated by the phase, width, and amplitude bits, respectively, provided to transmitter 540 for a given symbol period. Amplitude modulator 620 includes switches 740(a) and 740(b) which route START to edge-to-pulse generators (EPG) 730(a) and 730(b), respectively, depending on the state of the amplitude bit. Switches 740 may be AND gates, for example. _STOP is provided to second inputs of EPGs 730(a) and 730(b) (generically, EPG 730). On receipt of START, EPG 730 initiates a symbol pulse, which it terminates on receipt of _STOP. Depending on which EPG 730 is activated, a positive or a negative going pulse is provided to the output of transmitter 540 via differential output buffer 610.
Strobe transmitter 790 includes DM 750 and matching logic block 780. DM 750 delays CLK_PULSE to provide a strobe signal that is suitable for resolving the data phase choices p0 and p1 of symbol 420. For one embodiment of strobe transmitter 790, DM 750 positions the strobe evenly between the phase bit states represented by p0 and p1 (
In general, DM 750 and matching logic block 780 duplicate for the strobe the operations of transmitter 540 on data signals at the level of physical layout. Consequently, this delay matching is robust to variations in process, temperature, voltage, etc. In addition, the remainder of the communication channel from the output of transmitter 540, through board traces, electromagnetic coupler 240, board traces on the other side of coupler 240, and to the inputs of receiver 530 at the receiving device, may be matched in delays between data and strobe in order to keep the chosen relative timing. However, the matching of delays is one embodiment described for illustrative purposes and is not necessary to practice this invention. For example, if the circuits and remainder of the channel do not maintain matched data to strobe delays, receivers may calibrate for the relative timing of the strobe or even compensate for the absence of a strobe by recovering the timing from appropriately encoded data.
For a given symbol pulse, START may be deasserted (negative-going edge) before or after the corresponding _STOP is asserted. For example, the disclosed embodiment of transmitter 540 is timed with CLK_PULSE, and higher symbol densities may be obtained by employing narrow CLK_PULSEs. The widths of START and _STOP are thus a function of the CLK_PULSE width, while the separation between START and _STOP is a function of the width bits. The different possible relative arrivals of the end of START and beginning of _STOP may adversely impact the modulation of symbol 420 by the width bits. Specifically, transistor 734 may be on or off when a negative-going edge of _STOP terminates the symbol pulse. Node N may thus either be exposed to the parasitic capacitances at node P through transistor 734, or not. This variability may affect the delay of the trailing symbol edge through EPG 730 in an unintended way.
The disclosed embodiment of receiver 530 includes differential to single-ended amplifiers 920(a) and 920(b) which compensate for the energy attenuation associated with electromagnetic coupler 240. Amplifiers 920(a) and 920(b) produce digital pulses in response to either positive or negative pulses on the transferred signal (TR_SYMBOL in
Matching strobe receiver 902 similarly amplifies the accompanying differential strobe signal. For the disclosed embodiment, the received strobe is used to decode phase information in data symbol 420. Strobe receiver 902 includes differential to single-ended amplifiers 920(c) and 920(d) and matched circuitry 904. Matched circuitry 904 replicates much of the remaining circuitry in receiver 530 to match delays for data and strobe signals, similar to the matching of transmitter 540 and strobe transmitter 790. One embodiment of strobe receiver 902 includes circuits that correspond to phase demodulator 670 and width demodulator 680 with some minor modifications. For example, strobe buffer 990 buffers the received strobe for distribution to multiple receivers 530, up to the number of channels in, e.g., bus 210. Strobe buffer 990 may be large, depending on the number of receivers it drives. Data buffer 980 corresponds to strobe buffer 990. To save area, data buffer 980 need not be an exact replica of strobe buffer 990. The delays can also be matched by scaling down both data buffer 980 and its loading proportionately, relative to their counterparts in strobe receiver 902.
Uni-OR gate (UOR) 940(a) combines the outputs of amplifiers 920(a) and 920(b) to recover the first edge of TR_SYMBOL. The name uni-OR indicates that the propagation delay through gate 940 is uniform with respect to the two inputs. An embodiment of UOR 940 is shown in
The disclosed embodiment of phase demodulator 670 includes an arbiter 950(b) (generically, “arbiter 950”) and data buffer 980. Arbiter 950(b) compares the first edge recovered from the transferred symbol by UOR 940(a) with the corresponding edge from the recovered strobe by UOR 940(b), respectively, and sets a phase bit according to whether the recovered first edge of the symbol leads or follows the first edge of the strobe. An embodiment of arbiter 950 is shown in
The disclosed embodiment of amplitude demodulator 660 includes an arbiter 950(a) which receives the amplified transferred signals from amplifiers 920(a) and 920(b). Arbiter 950(a) sets an amplitude bit according to whether the output of amplifier 920(a) or 920(b) pulses first.
The disclosed embodiment of width demodulator 680 includes delay modules (DMs) 910, 912, 914, arbiters 950(c), 950(d), 950(e), and decoding logic 960. The recovered first symbol edge is sent through DMs 910, 912, and 914 to generate a series of edge signals having delays that replicate the delays associated with different symbol widths. For one embodiment of the invention, DMs 910, 912, and 914 may be implemented as programmable delay modules (
Latches 970(a), 970(b), 970(c), and 970(d) receive first and second width bits, the phase bit, and the amplitude bit, respectively, at their inputs, and transfer the extracted (inbound) bits to their outputs when clocked by a clocking signal. For the disclosed embodiment of receiver 530, the latches are clocked by sampling a signal from the delay chain of width demodulator 680 through the extra delay of DM 916. This latching synchronizes the demodulated bits to the accompanying strobe timing. In addition, a device 220 may require a further synchronization of the data to a local clock, e.g. clock synchronization circuit 560 in
The various components in an embodiment of interface 230 include a number of circuit elements that may be adjusted to compensate for process, voltage, temperature variations and the like. For example, compensation may entail adjusting the delay provided by a programmable delay module (DM 770), the gain provided by an amplifier (amplifier 920), or the termination resistance (device sets 690(a) and 690(b)).
The same calibration control settings can be distributed to DMs used throughout interface 230, such as DM 712, DM 910, etc. The desired delays of DMs in interface 230 are achieved by selecting a number of programmable delay modules 770 for each such DM which have the same ratio to the total number of delay modules 770 included in all the DMs 1000 as the ratio of the desired delay to the clock period. For example, if there are twenty total delay modules 770 in the sum of DMs 1000, one can select a delay of one tenth of the clock period by using two delay modules 770 for any particular DM used in interface 230. In addition, one can also choose a fractional extra delay for any particular DM by inserting small extra loads at the outputs of selected delay modules 770 which constitute that DM.
The calibration information obtained by calibration circuit 520 may also be used to control other circuit parameters, in the face of variable conditions. These may include the resistance of termination device 690 and gain of amplifier 920. This may be done by correlating the information contained in the delay control setting with the effects of like process, temperature, voltage, and like conditions on the other circuit parameters.
As described in the multi-drop bus system example above, the devices 220(2)–220(m) and the device 220(1) can communicate with each other across the bus 210. For simplicity in the following discussion, the devices 220(2)–220(m) will be referred to as slave devices (“slaves”) and the device 220(1) as a master device (“master”).
In this multi-drop bus environment, the master presents its clock (called the master clock, the core clock, or the system clock) to each of the slaves along the bus. The slaves each use the master clock to clock returned data signals back to the master. The slaves may send back the master clock (their version of the master clock as received by each slave at its bus position) with each returned data signal as an accompanying return clock (also called a forwarded clock). Alternately, an implicit return clock may be recovered by the master from the timing of the return data itself.
However, because of delays (e.g., propagation delays) on the bus and in slave circuitry, the received data signal and accompanying return clock may not be synchronized with the master clock. Thus, the master may not be able to process the data without resynchronizing it to the master clock. Because each of the slaves communicates with the master over the same bus (in some applications alternating from slave to slave in short bursts), and since different slaves may experience different delays with respect to the master clock, the master cannot lock onto a fairly steady and long-lived succession of return clock pulses. The master may need to know the return delay of data, including resynchronization, not just quickly, but a-priori, for example because a bus controller's scheduling algorithm may plan bus usage in advance.
The synchronization circuit 560 may be used to measure and record the phase information of each return clock required for resynchronization during a calibration time. This phase information can be used both by a bus scheduling algorithm and synchronization circuitry at run time, based on the address of the active slave, on a cycle by cycle basis. First, the master need only recover clock phase information from the return clock, as the clock's frequency is known (since it is inherited from the master clock sent to each slave to begin with). Second, the accuracy of the calibrated and recorded phase information used for resynchronization may be reduced by splitting the uses of the return clock into two categories: demodulation and resynchronization.
By guaranteeing that the data and the return clock from a slave undergo the same phase delays (within certain tolerances) from the slave to the master, the data and the return clock remain synchronous with respect to each other even though they are asynchronous with respect to the master. This guarantee can be made through circuitry such as the delay matching clock circuitry in
Furthermore, while circuitry included in the master performs the demodulation, the master need not be aware of the timing through this circuitry from cycle to cycle. For example, the master receivers are bisected into two separate clock domains that operate asynchronously. Demodulation follows suit after whatever timing is provided, cycle by cycle, by the accompanying return clock, while resynchronization operates according to the master clock. This reduces the requirements on resynchronization since demodulation carries most of the accuracy burden. For example, with 1.6 Gigabits per second test chips, modulation timing accuracy ranges in the tens of picoseconds. Resynchronization accuracy, dictated by such factors as clock jitter and latch setup and hold times, may be on the order of hundreds of picoseconds.
The demodulation treatment of the forwarded clock (actually, both forwarded clocks, the clock from the master to the slaves and from the slaves to the master) is shown in
Resynchronization of the return data involves resynchronizing the now fully digital (demodulated) signals from the timing of the forwarded clock, which may vary every cycle, to the master clock for handoff to the master's core circuitry. This resynchronizing entails avoiding a danger window of setup and hold times when latching the demodulated signals. The danger window generally represents time preceding or following the system clock in which sampled data should not be switching. Avoiding latch setup/hold violations, with as short a latency as possible and so that an entire bus width of data stays lined up within the same clock cycle, requires relatively loose timing accuracy, e.g., a few hundred picoseconds. This relatively loose timing accuracy means that digital calibration data for the relative phase of each slave timing uses few bits of measurement, storage, and control and is fairly robust to changing conditions (e.g., temperature, voltage shift, AC supply noise, etc.) from calibration to runtime.
Note that the resynchronization could be performed before the demodulation. This order of operation, however, would require some non-trivial adjustments before or during resynchronization to preserve the analog information present in the received data signals.
A function of resynchronization is to retime demodulated data so that no part of the master window 1102 overlaps with the slave window 1110 for a particular strobe. Retiming to prevent overlap between the windows 1102 and 1110 reduces or eliminates the chance of some data being sampled during switching, which could result in incorrect latching and/or metastable behavior in latching circuits. For example, data corresponding to waveforms (STROBEs) 1104 and 1106 in
The measurement circuit 1200 includes first and second arbiters 1202 a and 1202 b that detect if the strobe is not adequately after or not adequately before the system clock, respectively. The first arbiter 1202 a compares the strobe to a version of the system clock delayed by a first delay element 1204. The first delay element 1204 is chosen as the sum of part of the master window 1102 (e.g., right part 1112 in waveform 1100 in
If the strobe received at the master from any slave is in the danger window (i.e., is neither adequately before or after the system clock), then calibration memory 1206 records the necessity for retiming the data from that slave. The master reapplies the slave address for that slave at run time and retrieves and distributes the retiming bit to data synchronization blocks on a per clock cycle basis.
An example of such a data synchronization block is the compensation circuit 1300 of
If the strobe for the currently active slave was in the danger window then the data is delayed through a delay element 1304 to avoid the danger window. The delay through the delay element 1304 should be larger than the sums of the master window 1102 and the slave window 1110 but small enough to not retime the data a full cycle and back into another danger window. Because the sum of the master window 1102 and the slave window 1110 is likely a fraction of a clock period, calculating the delay should not require much accuracy. If more accuracy is desired, however, it can be obtained by using more bits of calibration. Using more bits of calibration can provide more information about where the data lies in relation to the danger window and therefore more information on how to accurately avoid the danger window. Delay element 1304 may include both calibrated and uncalibrated components.
Calibrating for bus return time is not limited to implementation in a multi-drop bus scenario as described above. Any system or circuitry having a device or mechanism receiving data from multiple mechanisms or devices across links with varying delays may be calibrated according to circuits such as the measurement circuit 1200 and the compensation circuit 1300 or 1400.
There has thus been disclosed a mechanism for calibrating return time, such as in the context of high bandwidth communications in multi-drop bus systems.
Other embodiments are within the scope of the following claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US6029250||Sep 9, 1998||Feb 22, 2000||Micron Technology, Inc.||Method and apparatus for adaptively adjusting the timing offset between a clock signal and digital signals transmitted coincident with that clock signal, and memory device and system using same|
|US6340901 *||Mar 9, 2000||Jan 22, 2002||Sun Microsystems, Inc.||Measurement of signal propagation delay using arbiters|
|US6581017||Jun 28, 2001||Jun 17, 2003||Intel Corporation||System and method for minimizing delay variation in double data rate strobes|
|US6643787||Oct 19, 1999||Nov 4, 2003||Rambus Inc.||Bus system optimization|
|US6665624||Mar 2, 2001||Dec 16, 2003||Intel Corporation||Generating and using calibration information|
|1||*||C.E. Molnar & I.W. Jones, "Simple circuits that work for complicated reasons," Proceedings of the Sixth International Symposium on Advanced Research in Asynchronous Circuits and Systems, IEEE, pp. 138-149, Apr. 2000.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7225283 *||Dec 21, 2004||May 29, 2007||Cypress Semiconductor Corporation||Asynchronous arbiter with bounded resolution time and predictable output state|
|US7383370||Mar 20, 2006||Jun 3, 2008||Cypress Semiconductor Corporation||Arbiter circuit and signal arbitration method|
|US7821301 *||Jan 11, 2006||Oct 26, 2010||International Business Machines Corporation||Method and apparatus for measuring and compensating for static phase error in phase locked loops|
|US8008907 *||Oct 12, 2006||Aug 30, 2011||Infineon Technologies Ag||Delay line calibration circuit comprising asynchronous arbiter element|
|US20070159213 *||Jan 11, 2006||Jul 12, 2007||Jenkins Keith A||Method and apparatus for measuring and compensating for static phase error in phase locked loops|
|US20070182423 *||Oct 12, 2006||Aug 9, 2007||Infineon Technologies Ag||Delay line calibration circuit comprising asynchronous arbiter element|
|U.S. Classification||713/400, 327/19, 327/3, 713/401|
|International Classification||G06F1/12, H04J3/06, H04L7/00|
|Cooperative Classification||H04L7/0008, H04J3/0682|
|Oct 22, 2004||AS||Assignment|
Owner name: INTEL CORPORATION, CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:SIMON, THOMAS D.;AMIRTHARAJAH, RAJEEVAN;REEL/FRAME:015283/0991
Effective date: 20010226
Owner name: INTEL CORPORATION, CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:SIMON, THOMAS D.;AMIRTHARAJAH, RAJEEVAN;REEL/FRAME:015284/0030
Effective date: 20010226
|Dec 7, 2009||REMI||Maintenance fee reminder mailed|
|May 2, 2010||LAPS||Lapse for failure to pay maintenance fees|
|Jun 22, 2010||FP||Expired due to failure to pay maintenance fee|
Effective date: 20100502